Authors: Xue Geng, Zhe Wang, Chunyun Chen, Qing Xu, Kaixin Xu, Chao Jin, Manas Gupta, Xulei Yang, Zhenghua Chen, Mohamed M. Sabry Aly, Jie Lin, Min Wu, Xiaoli Li

Published on: May 09, 2024

Impact Score: 8.0

Arxiv code: Arxiv:2405.06038

Summary

• What is new: This research provides a comprehensive survey of recent advancements in model compression techniques, hardware accelerators for DNNs, and integration of homomorphic encryption for security. It provides an overarching view from algorithm, hardware, to security aspects for efficient DNN deployment.
• Why this is important: Deploying deep neural networks is costly in terms of memory, energy, and computation, along with the need to maintain security and privacy.
• What the research proposes: A detailed survey of model compression methods, advancements in custom DNN hardware accelerators, and the use of homomorphic encryption to enhance security and privacy in DNN deployment.
• Results: The survey synthesizes a vast array of research into a coherent overview of techniques for efficient, secure DNN deployment, identifying key trends, challenges, and potential directions.

Technical Details

Technological frameworks used: Model quantization, Model pruning, Knowledge distillation, Optimizations of non-linear operations

Models used: Custom DNN hardware accelerators

Data used: Homomorphic encryption for secure DNN deployment

Potential Impact

Tech companies deploying DNNs, AI-focused startups, companies in the AI hardware accelerator market, and industries concerned with AI security and privacy.

Want to implement this idea in a business?

We have generated a startup concept here: EffiNetAI.